The Effects of Credit Competition on Banks’ Loan Loss Provision Timeliness

Yiwei Dou Stern School of Business

New York University ydou@stern.nyu.edu

Stephen G. Ryan

Stern School of Business New York University sryan@stern.nyu.edu

Youli Zou

Rotman School of Management University of Toronto

youli.zou08@rotman.utoronto.ca

March 1, 2013

We thank Tara Rice for providing her data on state-level interstate branching restrictions and Allan Collard-Wexler for providing his contiguous county data. We thank Jeff Callen and Alvis Lo for comments on an earlier draft, Dick Sylla and Larry White for insights regarding the historical development of bank regulation, and Scott Liao and seminar participants at New York University for useful comments.

The Effects of Credit Competition on Banks’ Loan Loss Provision Timeliness

Abstract: Prior research has documented specific benefits, but not costs, of timelier loan loss provisioning for reporting banks and the overall financial system. We examine one such cost in this paper. We argue that potential entrants into a local loan origination market use incumbent banks’ reported loan loss provisions to assess incumbents’ loan underwriting quality and thus the desirability of market entry. By loan underwriting quality, we mean banks’ ability to evaluate credit risk to determine which loans to grant as well as the interest rates and other contractual terms of granted loans. Incumbents with better loan underwriting quality leave fewer profitable lending opportunities for potential entrants. We use variation in interstate branching deregulation across contiguous counties of adjacent states from 1994 to 2005 as a natural experiment to investigate how increased threat of entry affects incumbent banks’ loan loss provision timeliness. We predict and find that incumbents more subject to entry reduce the timeliness of their bad news (but not good news) loan loss provisions to increase their perceived loan underwriting quality. Further, we predict and find variation in this behavior across loan origination markets attributable to incumbents’ loan portfolio composition and borrower turnover in those markets. Keywords: banks, deregulation, interstate branching, credit competition, loan loss provision, timeliness, natural experiment

1  

1. Introduction

Some of the most influential financial policymakers emphasize the benefits of timelier

loan loss provisioning for both reporting banks and the overall financial system. These

policymakers typically argue that FAS 5’s incurred loss model—which allows banks to

recognize loan losses only if they are “incurred”, “probable”, and “can be reasonably

estimated”—delays loan loss provisions during good economic times, rendering banks less well

capitalized and thus less willing to originate loans when the business cycle turns down (Bernanke

2009; Dugan 2009; Financial Stability Forum 2009; U.S. Treasury 2009). The prior literature

generally supports this argument.1 Beatty and Liao (2011) and Bhat et al. (2012) find that banks

that record timelier loan loss provisions exhibit lower loan origination procyclicality, because

they maintain higher common equity during good economic times and understand their credit

risk better. Bushman and Williams (2012a,b) find that these banks assume lower financial

leverage, engage in less risk shifting (exploitation of deposit insurance), and are less prone to

experience balance sheet contraction and to contribute to system-wide contraction, because

timelier provisions increase bank transparency, enhance market discipline, and reduce

unrecognized loss overhang.

While prior research has documented these and other specific benefits of timelier loan

loss provisioning to banks and the overall financial system, it has not documented specific actual

or perceived costs of such provisioning to banks.2 We examine one such cost in this paper. We

argue that potential entrants into a loan origination market use incumbent banks’ reported loan

                                                            1 Prior papers typically identify timelier loan loss provisions as provisions that are more strongly positively associated with future and sometimes also contemporaneous changes in non-performing loans. As discussed below, we identify timelier loan loss provisions consistently as annual provisions that are more positively associated with the next-year change in non-performing loans. 2 Considerable prior research, most of which is conducted on samples of firms excluding financial institutions, explains that timelier recognition of losses (of any type, not just provisions for loan losses) may entail contracting and other costs; see Gormley et al. (2012) for a recent example. 

2  

loss provisions to assess the incumbents’ loan underwriting quality and thus the desirability of

market entry. By loan underwriting quality, we mean banks’ ability to evaluate credit risk to

determine which loans to grant as well as the interest rates and other contractual terms of granted

loans. Incumbent banks with better loan underwriting quality leave fewer profitable lending

opportunities for potential entrants. We use variation in interstate branching deregulation across

contiguous counties of adjacent states from 1994 to 2005 to investigate how increased threat of

entry affects incumbents’ loan loss provision timeliness.

Prior research shows that information asymmetry between banks offering credit and

potential borrowers affects banking industry structure and conduct (Dell’Ariccia et al. 1999;

Marquez 2002; Dell’Ariccia and Marquez 2004, 2006, 2008). Due to their greater local market

knowledge, including established relationships with borrowers, incumbent banks are less subject

than potential entrants to this form of information asymmetry. This information advantage

provides incumbents with market power. In extreme cases, fear of the “winner’s curse” renders

potential entrants unwilling to provide credit to incumbent banks’ borrowers, so that incumbent

banks capture those borrowers (Rajan 1992; Dell’Ariccia and Marquez 2004).

We develop our hypotheses based on three interrelated arguments. First, potential

entrants into a loan origination market can use incumbent banks’ loan loss provisions to infer

their loan underwriting quality. All else being equal, an incumbent that experiences higher

economic loan losses evidences lower loan underwriting quality. That incumbent is more likely

to be subject to entry when these losses are observable to a greater extent by potential entrants.

Second, incumbent banks can manage this extent because potential entrants cannot perfectly

distinguish timelier increases in loan loss provisions (hereafter, “bad news loan loss provisions”)

from higher economic loan losses, despite the existence of other information about economic

3  

loan losses. We validate the first two arguments, showing that incumbent banks that record

timelier bad news (but not good news) loan loss provisions experience greater entry. Third, an

incumbent bank’s decisions regarding the timeliness of its loan loss provisions reflect a trade-off

between the decision-making benefits described above and competitive costs. These costs

increase with the threat of entry, tilting incumbent banks’ incentives toward recording less timely

bad news provisions.

While prior theoretical research shows that the threat of entry affects incumbent firms’

financial reporting decisions (Darrough and Stoughton 1990), documenting this relation

empirically requires confronting at least three research-design challenges. First, entry and

incumbents’ financial reporting decisions likely are endogenously determined. Second, firms

often comprise multiple business units competing in different local markets, and relevant data to

assess this relation are usually not publicly available for these units. Third, while it is possible to

control for observable factors that affect incumbents’ reporting decisions, unobservable factors

also likely play sizeable roles in these decisions. Our research design addresses these challenges.

We address the first challenge by examining the effect of increased threat of entry by out-

of-state banks caused by state-level interstate branching deregulation from 1994 to 2005 on

incumbent banks’ loan loss provision timeliness. Individual states implemented this deregulation

to varying extents and at different times during this period. While this deregulation was shaped

by a political process involving the relative influence of small versus large banks and other

factors (Strahan 2003; Rice and Strahan 2010), this process has no apparent relation to

incumbent banks’ preexisting or intended loan loss provision timeliness at the time, and so we

argue it constitutes a natural experiment for the purposes of this study.

4  

Specifically, the Riegle-Neal Interstate Banking and Branching Efficiency Act of 1994

(IBBEA) eliminated federal restrictions on interstate banking (out-of-state bank acquisition or

creation of in-state chartered banks) and branching (out-of-state bank acquisition or creation of

in-state branches) expressed in the McFadden Act of 1927 and the Bank Holding Company Act

of 1956 (Johnson and Rice 2008). However, IBBEA’s provisions allowed states to continue to

restrict interstate branching, which is a cheaper and easier means of entry than interstate banking.

Some states imposed one or more of: (1) a minimum age at which in-state banks can be acquired

by out-of-state banks, (2) a statewide cap on the amount of deposits that the merged institution

could hold after an out-of-state bank acquires an in-state bank, (3) a prohibition on the creation

of new branches by out-of-state banks (de novo branching), and (4) a prohibition on the

acquisition of single branches by out-of-state banks. Collectively, these restrictions significantly

reduced entry by out-of-state banks (Johnson and Rice 2008).

Following Rice and Strahan (2010), we construct an index of interstate branching

restrictions for each state. The index is zero for states without entry restrictions (the greatest

threat of entry) and increases by one for each of the four restrictions described above, up to a

maximum of four (the least threat of entry).

We overcome the second and third challenges by using branch addresses obtained from

the Federal Deposit Insurance Corporation (FDIC) to identify banks with branches in only one

county (“one-county banks”); these local banks also are relatively comparable in their (small)

size. We compare one-county banks in states that have experienced deregulatory events to

matched one-county banks in contiguous counties of adjacent states that have experienced

strictly less deregulation. This matched-pair research design both minimizes differences in

unobservable factors across banks and enhances our ability to control for observable factors.

5  

Given the benefits of timelier loan loss provisions documented by the prior literature, we

expect incumbent banks to reduce the timeliness of their bad news provisions to dissuade entry

only when potential entrants rely to a significant extent on those provisions to assess the returns

to entering the incumbents’ loan origination markets.3 We expect this to occur primarily for

heterogeneous (e.g., commercial and industrial) loans, for which relatively little information

about the loans’ creditworthiness is available outside of incumbent banks’ financial and

regulatory reports, local market knowledge is important to originate the loans effectively, and the

winner’s curse problem is significant. We also expect this to be more the case when the loan

origination market exhibits lower borrower turnover.

Following extensive prior literature (Gambera 2000; Nichols et al. 2009; Beatty and Liao

2011; Bhat et al. 2012; Bushman and Williams 2012a,b), we estimate banks’ loan loss provision

timeliness by regressing annual loan loss provisions on next-year changes in non-performing

loans and other explanatory variables. In this approach, changes in non-performing loans proxy

for economic credit losses, and a more positive coefficient on next-year changes in

nonperforming loans indicates timelier loan loss provisions. The prior literature finds that loan

loss provisions predict next-year changes in non-performing loans for the average bank, with

considerable variation in this predictive relation obtaining across banks. These findings reflect

the fact that banks can base their provisions on all available information, including loans that are

delinquent but not yet non-performing. Because we are primarily concerned with the timeliness

of bad news loan loss provisions, we generalize the usual approach to allow the coefficient on

                                                            3 Prior research that examines the effects of conservatism in contracting and market settings typically posits that firms must commit to conservatism to obtain the desired outcomes (e.g., improved contracting or lower cost of capital). In contrast, we posit only that potential entrants place some weight on incumbent banks’ loan loss provisions in assessing their economic loan losses and thus loan underwriting quality.

6  

next-year changes in non-performing loans to vary with the direction of the change, and focus on

the coefficient on the increase in non-performing loans.

Consistent with the expectations discussed above, we find that the timeliness of bad news

loan loss provisions, but not good news provisions, decreases for incumbent banks located in the

counties where restrictions on interstate branching decrease. Moreover, we find the reduction of

loan loss provision timeliness is higher for incumbents in counties where banks hold a larger

proportion of heterogeneous loans and that exhibit lower borrower turnover. Our findings are

robust to numerous specification analyses.

Our paper contributes to several extant literatures. First, it informs the ongoing policy

debate about the desirability of requiring banks to record timelier loan loss provisions than they

currently report under FAS 5’s incurred loan loss model. Our findings suggest that imposing this

requirement would lead to greater entry into local loan origination markets.

Second, our paper contributes to the literature on the competitive costs of financial report

information. Theoretical research demonstrates that the relation between competition and

disclosure depends on the natures of both variables (Darrough 1993; Verrecchia 1990).

Intuitively, an incumbent firm’s disclosures can deter entry by conveying either that the firm is

strong (Milgrom and Roberts 1982) or that the market is undesirable (Darrough and Stoughton

1990). In cases where disclosure would induce entry, firms do not voluntarily disclose (Pagano

and Jappelli 1993). We provide evidence that incumbent banks exposed to greater threat of entry

reduce the timeliness of their bad news loan loss provisions to appear to have higher loan

underwriting quality and thus deter entry.

Our paper is related to two recent studies. Burks et al. (2012) employ the same setting

and find that banks exposed to greater threat of entry issue more press releases. They do not

7  

provide evidence as to whether these press releases tend to convey good or bad news or whether

good and bad news releases have different implications for entry, however. Bushman et al.

(2013) show that a measure of banks’ perceived competition derived from textual analysis of

their financial reports is negatively associated with banks’ loan loss provision timeliness, a

finding the authors attribute to banks in competitive (i.e., low-profit) environments delaying

provisions to appear to be more profitable rather than to have better loan underwriting quality.

Third, our paper contributes to the literature documenting the beneficial effects of

conditional conservatism. This literature generally finds that firms’ commitment to timely loss

recognition increases contracting efficiency and reduces information asymmetry in capital

markets (Kothari et al. 2010). Our evidence indicates that timelier bad news loan loss provisions

by incumbent banks increases entry into their loan origination markets. While costly for the

reporting banks, entry likely enhances social welfare, because the prior literature finds that credit

competition improves bank services (Jayaratne and Strahan 1998; Dick 2006), expands credit

availability and lowers interest rates (Rice and Strahan 2010), and increases local economic

growth (Jayaratne and Strahan 1996).

Finally, our research extends the industrial organization literature examining factors that

create barriers to entry in markets. This paper demonstrates that incumbents’ strategic accounting

choices play a role in creating these barriers. Our paper is related to Zou (2012), who examines

how entrant airlines use incumbent airlines’ accumulated other comprehensive income arising

from cash flow hedges of fuel costs in making route entry decisions.

The remainder of the paper is organized as follows. Section 2 provides relevant

background and develops the hypotheses. Section 3 describes the sample and research design.

8  

Section 4 discusses the primary empirical results and Section 5 summarizes the robustness

analyses performed. Section 6 concludes.

2. Background and Hypothesis Development

2.1. Means of Geographical Expansion by Banks

Absent the restrictions discussed below, an out-of-state bank can expand geographically

into a state in two primary ways. First, it can charter a new bank in the state or acquire an

existing chartered bank in the state and maintain the bank’s charter. These activities are referred

to as “interstate banking.” Second, it can acquire an existing bank in the state and convert the

bank to one or more branches or establish new branches in the state. These activities are referred

to as “interstate branching.” Interstate banking is a significantly more expensive means to

expand geographically than interstate branching, because chartered banks require separate

management and governance and must meet regulatory capital and other requirements on a

standalone basis.

Banks operating in a state can expand geographically by acquiring or opening new

branches in the state. These activities are referred to as “intrastate branching.”

2.2. Relaxation of Restrictions on Geographical Expansion by Banks

State-imposed restrictions on banks’ geographical expansion arose early in United States

history, in part as a consequence of provisions of the U.S. Constitution (Kroszner and Strahan

1999).4 Beginning in the late 19th century and accelerating in the early 20th century, national

                                                            4 The U.S. Constitution prohibited states from issuing fiat (i.e., paper) money and taxing interstate commerce. States imposed fees on bank charters and banking-related taxes and invested in banks to produce a substantial portion of their total revenues (Sylla et al. 1987). Because states could not earn these revenues on national banks, they prevented these banks from expanding into their territories.  

9  

banks developed first multi-bank chains (chartered banks in different states owned by one or a

few individuals) and then multi-bank holding companies (entities that own separately chartered

banks in different states) to evade these restrictions (Redlich 1951). While nominally leveling

the competitive playing field for the state and national banks operating in a state, the McFadden

Act of 1927 effectively reaffirmed the states’ authority over branching by national banks in their

territories. The McFadden Act did not restrict multi-bank holding companies, however, inducing

further expansion of these companies, notably A. P. Giannini’s founding of Transamerica

Corporation in 1928 and Bank of America in 1930. The Bank Holding Company Act of 1956

gave states the authority to restrict interstate banking by these holding companies, thereby

returning to these states the power to restrict expansion by out-of-state banks. Prior to 1970, all

states prohibited interstate branching, and they significantly restricted interstate banking and

intrastate branching.

Most states deregulated interstate banking and intrastate branching from 1970 to 1994

due to developments in technology (e.g., the ATM), competition (e.g., non-banks offering

banking services such as cash management accounts), and other factors (Berger et al. 1995;

Kroszner and Strahan 1999; Ryan 2007). This deregulation typically occurred on a gradual basis

through reciprocal agreements with other states. In contrast, restrictions on interstate branching

remained until the passage of IBBEA in 1994.

IBBEA permitted interstate banking (Section 101) and branching (Sections 102 and 103).

Section 101 had relatively modest effects due to the prior deregulation of interstate banking. In

contrast, Sections 102 and 103 permitted interstate branching for the first time, significantly

reducing the cost to out-of-state banks of entering a state (Rice and Strahan 2010).

10  

IBBEA provided states with various means to control the extent and timing of the

implementation of the act’s interstate branching provisions, however. States could “opt-in” early

or “opt-out” of these provisions by passing a state law at any time between IBBEA’s passage in

September 1994 and the “trigger date” of June 1, 1997. All fifty states and Washington D.C.

eventually opted in, although this occurred after considerable debate in nine states, including

initial votes to opt out in three states (Golembe 1994; Hutnyan 1995). As discussed below, states

could and to varying extents did impose four restrictions on interstate branching. Collectively,

these restrictions erected barriers to entry by out-of-state banks, raised the cost of entry, and

distorted the means of entry.

The first restriction is imposition of the minimum age at which in-state banks can be

acquired by out-of-state banks. IBBEA allows states to set a minimum up to five years. This

restriction may prolong the process of or even deter entry, depending on out-of-state banks’

patience and opportunities to acquire sufficiently old in-state banks.

The second restriction is imposition of statewide deposit caps on the merged institutions

created by acquisitions of in-state banks by out-of-state banks. IBBEA specifies a statewide

deposit cap of 30%, but a state may impose a cap below 30%. This restriction prohibits any

interstate acquisition for which the deposits of the merged institution exceed the specified cap.

The third restriction is prohibition of the creation of new branches by out-of-state banks.

IBBEA allows such de novo interstate branching only if state law expressly permits it. This

restriction makes entry both more expensive and less flexible, because entry is possible only via

acquisition of existing banks or branches.

The fourth restriction is prohibition of the acquisition of single branches by out-of state

banks. IBBEA states that an out-of-state bank may acquire branches of an in-state bank, without

11  

acquiring the entire bank, only if the state permits such acquisitions. Similar to the third

restriction, this restriction makes entry both more expensive and less flexible, because to acquire

desired branches out-of-state banks may also have to acquire undesired branches and other

portions of in-state banks.

States may amend their restrictions on interstate branching over time. Fifteen states

amended their initial restrictions after the 1997 trigger date.

Following Rice and Strahan (2010), we construct an index of these four restrictions on

interstate branching for each state in each year from 1994 to 2005. The index, denoted RESTR,

is the sum of four zero-one indicator variables: MINAGE equals one if a state imposes a

minimum age of three years or more on target institutions; DEPCAP equals one if a state

imposes a deposit cap of less than 30%; DENOVO equals one if a state does not permit de novo

interstate branching; and SGLBR equals one if a state prohibits the acquisition of single branches

by out-of-state banks. RESTR ranges from zero (the greatest threat of entry) to four (the least

threat of entry).5

Strahan (2003) and Rice and Strahan (2010) argue that differential deregulation on

interstate branching across states, while shaped by a political process involving the relative

influence of small versus large banks and other factors, represents an “ideal laboratory” or

“instrument” to identify the effects of changes in banks’ competitive environment. For example,

Rice and Strahan provide evidence that this process was uncorrelated with the demand for credit.

This setting has been used in numerous studies to identify the effect of competition on banking

market structure (Dick 2006; Johnson and Rice 2008), banks’ cost or profit efficiency (Koetter et

                                                            5 Some states had not finalized their restrictions on interstate branching by 1994 or 1995. Following Tara Rice’s suggestion, we code the branching restriction index for these state-year observations as four, the least threat of entry.

12  

al. 2012), small business finance (Rice and Strahan 2010), and private firms’ borrowing and

investment (Zarutskie 2006).

None of the evidence in the literature suggests that the political process that led to these

deregulatory events was associated with incumbent banks’ pre-existing or intended loan loss

provision timeliness. For this reason, we argue these events constitute a natural experiment for

the purposes of this study. Nevertheless, we conduct robustness analysis showing that our

reported results are robust to controlling for six variables affecting the political process identified

by Rice and Strahan (2010), as well as related robustness analyses described in Section 5.4.

2.3. Loan Loss Provisions

An extensive empirical literature documents that banks’ loan loss provisions, along with

their supplemental financial and regulatory report disclosures, provide information to investors

about the performance and creditworthiness of banks’ loan portfolios (Wahlen 1994; Liu and

Ryan 1995; Liu et al. 1997; Docking et al. 1997; Ahmed et al. 1999; Kanagaretnam et al. 2004).

These findings obtain despite the fact that FAS 5’s incurred loss model—which allows banks to

recognize loan losses only if they are “incurred”, “probable”, and “can be reasonably

estimated”—delays loan loss provisions to greater or lesser extent depending on the type of loan

and other factors. See Ryan (2012a, Section 3) for a recent survey of this empirical literature and

extensive discussion of FAS 5.

This literature also shows that loan loss provisions have a strong positive association with

future changes in non-performing loans for the average bank, indicating that these provisions are

a relatively timely measure of credit losses on loans, and documenting significant variation in

loan loss provision timeliness across banks and time (Gambera 2000; Nichols et al. 2009; Beatty

13  

and Liao 2011; Bhat et al. 2012; Bushman and Williams 2012a,b; Bushman et al. 2013). This

literature further finds that banks that record timelier loan loss provisions generate various

specific benefits for themselves and the overall financial system: less pro-cyclical lending

(Beatty and Liao 2011; Bhat et al. 2012), lower leverage, less risk-shifting (exploitation of

deposit insurance), less exposure to balance sheet contraction, and less contribution to system-

wide balance-sheet contraction (Bushman and Williams 2012a,b).

Given these benefits, the existence of significant variation in the timeliness of loan loss

provisions across banks and time suggests that banks perceive timely loan loss provisioning to

involve costs. We examine one specific cost in this study—potential entrants may use

incumbent banks’ loan loss provisions to make inferences about their loan underwriting quality

in deciding whether to enter the incumbents’ markets.

2.4. Hypothesis Development

Although credit histories, scores, or ratings are available for many consumer and

corporate borrowers, a pervasive problem in banking is that potential borrowers know more

about their current creditworthiness and future prospects than do lenders. Moreover, the extent

of this information asymmetry varies across banks, with incumbent banks being more

knowledgeable than potential entrants.   For example, loan agreements usually grant existing

lenders the right to inspect borrowers’ properties and books (Taylor and Sansone 2007), and

checking account agreements usually grant these lenders the right to receive detailed accounts

receivable and inventory information (Mester et al. 2007). Even ignoring their existing customer

relationships, incumbents have greater local market knowledge. This knowledge provides

incumbents with insight into the credit risks of loan applicants, enabling them to make better

14  

decisions regarding whether and at what interest rate and other terms to grant credit than can

potential entrants (Gan and Riddiough 2008). Potential entrants generally have to incur

significant costs to acquire customer relationships or local market knowledge. Berger and Dick

(2007) provide evidence that entrants invest heavily in branch networks to gain access to

borrowers, depositors, and local market knowledge, and that these investments cause entrants’

information disadvantage to decline gradually over time.

Prior research shows that incumbent banks’ information advantage over potential entrants

has significant effects on banking industry structure and conduct (Dell’Ariccia et al. 1999;

Marquez 2002; Dell’Ariccia and Marquez 2004, 2006, 2008). This advantage provides

incumbents with market power. In extreme cases, fear of the “winner’s curse” renders potential

entrants unwilling to provide credit to incumbents’ borrowers (Rajan 1992; Dell’Ariccia and

Marquez 2004).

Incumbent banks with better loan underwriting quality leave fewer profitable lending

opportunities for potential entrants. Hence, potential entrants should attempt to determine

whether incumbents’ loan underwriting quality is sufficiently low that entering the local loan

origination market would be profitable.

We argue that incumbent banks exercise discretion to reduce the timeliness of their bad

news provisions, but not their good news provisions, to appear to have good underwriting quality

and thereby dissuade potential entrants from entering their loan origination markets. Logically,

potential entrants might infer one of three primary non-discretionary drivers of incumbents’

provisions for loan losses, holding other information about incumbent banks’ economic loan

losses constant: (1) loan underwriting quality, (2) the arrival of news about loan performance

15  

after loan origination, and (3) FAS 5’s accounting requirements.6 The first two drivers pertain to

the level of incumbents’ economic loan losses and the third pertains to the timeliness of

incumbents’ reported loan loss provisions. Below, we describe the different inferences that

potential entrants likely would make about these drivers from bad versus good news provisions.

Potential entrants are likely to assign considerable probability that poor loan underwriting

quality causes bad news loan loss provisions, but are unlikely to assign much probability that

good underwriting quality causes good news provisions. The reason is it is far easier for

incumbents with poor loan underwriting quality to write loans for which they underestimate

credit losses than for either incumbents with poor (good) loan underwriting quality to write loans

for which they overestimate (accurately estimate) credit losses. This is the winner’s curse

problem mentioned above. Potential entrants likely assign similar probability that the arrival of

bad (good) news after loan origination causes bad (good) news loan provisions.

Potential entrants are unlikely to assign much probability that FAS 5 causes bad news

loan loss provisions, but may infer that FAS 5 causes good news loan loss provisions, because

the standard’s three conditions for recording losses require banks to compile sufficient evidence

of loan default before recording loan loss provisions. These conditions actively work against

banks recording timely bad news provisions, and may yield good news loan loss provisions.

Based on the discussion above, we expect potential entrants to ascribe bad news loan loss

provisions first to bad underwriting quality, second to bad luck, and third to FAS 5. We expect

potential entrants to ascribe good news loan loss provisions first to good luck, second to FAS 5,

and third to good loan underwriting quality. Consistent with these expectations, in a validation

                                                            6 Prior research documents numerous discretionary drivers of banks’ loan loss provisions and related inferences by market participants about banks’ health and other characteristics that we do not discuss here; see Ryan (2012a, Section 3.3) for a recent summary of this research.

16  

test we show that timelier bad news loan loss provisions lead to more entry but that timelier good

news loan loss provisions have no effect on entry.7

We argue that an incumbent bank’s decisions regarding the timeliness of its bad news

loan loss provisions should reflect a trade-off between the benefits associated with less

procyclical lending and risk-taking and the costs of encouraging more entry. The cost of

incurring more entry increases when deregulation intensifies entry threats, tilting incumbents’

incentives toward reducing the timeliness of their bad news loan loss provisions. This yields our

first hypothesis:

H1a: The timeliness of incumbents’ bad news loan loss provisions decreases with the threat of entry by out-of-state banks into incumbents’ local markets. H1b: The timeliness of incumbents’ good news loan loss provisions has no association with the threat of entry by out-of-state banks into incumbents’ local markets.

We refer to hypotheses H1a and H1b collectively as H1.

Given the benefits of timelier loan loss provisions documented by the prior literature, we

expect incumbent banks to reduce the timeliness of their provisions to deter entry only when

potential entrants rely to a significant extent on those provisions to assess the returns to entering

the incumbents’ local loan origination markets. We expect this to occur primarily for

heterogeneous (e.g., commercial and industrial) loans, for which relatively little information

about the loans’ creditworthiness is available outside of incumbent banks’ financial and

                                                            7 This finding rules out the possibility that potential entrants infer that incumbents have better post-origination credit risk modeling (Bhat et al. 2012) from timelier bad news loan loss provisions, which would discourage entry. Despite this fact, we control for size in our empirical models, because Bhat et al. find that bank size is the most significant determinant of the quality of credit risk modeling.

17  

regulatory reports, local market knowledge is important to originate the loans effectively, and the

winner’s curse problem is significant.8

We also expect this to be more the case when the local loan origination market exhibits

higher borrower turnover, which we proxy for using growth in non-government establishments.

Marquez (2002) demonstrates that the information asymmetry between incumbent and potential

entrant banks is mitigated by higher borrower turnover.

This discussion yields the following hypotheses:

H2a: Hypothesis H1a holds more strongly as the incumbent banks’ proportion of heterogeneous loans increases. H2b: Hypothesis H1a holds less strongly as borrower turnover in the local market increases.

We refer to hypotheses H2a and H2b collectively as H2.

3. Data and Research Design

3.1. Sample Selection and Matching of Treatment and Control Banks

Our sample selection is intended to address the three research-design challenges

described in the introduction by yielding samples of treatment and control banks that are

matched as closely as possible based on local markets and other characteristics but for which the

two samples are subject to strictly different levels of interstate branching deregulation. We refer

                                                            8 Alternatively, incumbents might believe that their information advantage for heterogeneous loans is so large that they are adequately protected against entry, and thus be less likely to defer bad news loan loss provisions. To rule out this possibility, for each county-year, we calculate the average proportion of commercial and industrial loans for all one-county banks, weighted by total assets, and next year’s proportion of branches owned by out-of-state banks. We find that these variables are significantly positively correlated, consistent with incumbents being more exposed to entry on heterogeneous than homogeneous loans.  

18  

to the sample of banks with less (more) restricted interstate branching, as indicated by lower

(higher) RESTR, as the treatment (control) sample.

We perform this matching at the county level, primarily because the banking literature

conventionally treats counties as local markets (Gilbert and Kochin 1989; Clair et al. 1994;

Prager and Hannan 1998; Berger et al. 1999; Rhoades 2000; Calomiris and Mason 2003;

Ashcraft 2005; Garmaise and Moskowitz 2006; Huang 2008).9 This convention is consistent

with the geographical proximity of banks to most of their borrowers and depositors. For

example, Petersen and Rajan (2002) report a median distance of five miles between small

business borrowers and their primary lending banks during 1990-1993. The counties included in

our sample have a median size of 587 square miles, so parties located five miles apart likely are

located in the same county. To obtain variation in deregulation of interstate branching, however,

we match contiguous counties of adjacent states with strictly different deregulation. Figure 1

depicts an example of such contiguous counties.

Tara Rice provided us with data on the four types of interstate branching restrictions by

state from 1994 to 2005. For each year in this period, we identify pairs of neighboring states for

which one state imposes all of the interstate branching restrictions that the other does and at least

one additional restriction; we refer to the restrictions in the former state as “strictly tighter”.10

For a pair of states to be included in the sample, we require both states to have the same

restrictions for at least one year during our sample period (we refer to these years as “pre-

deregulation”) and one state to have strictly tighter restrictions for at least three subsequent years

                                                            9 Similar local geographic-matching approaches have been used in other social science literatures (Fox 1986; Card and Kruger 1994; Holmes 1998; Black 1999; Huber and Arceneaux 2007; Keele and Titiunik 2012). 10 To illustrate, if state A imposes a minimum age of five years for in-state banks to be acquired by out-of-state banks and state B forbids de novo interstate branching and the acquisition of single branches of an institution, then we deem two states’ restrictions to be not comparable even though state B has more restrictions. In contrast, if state C has the same minimum age requirement as state A and prohibits de novo interstate branching, then state C’s restrictions are strictly tighter than those in state A.

19  

(we refer to these years as “post-deregulation”). The first part of this requirement mitigates the

heterogeneity of states in the pre-deregulation period. The second part provides a reasonable

period of time in the post-deregulation period for incumbent banks to be subject to increased

threat of entry by out-of-state banks and to adjust their loan loss provision timeliness accordingly.

Table 1 lists the 68 state pairs meeting this requirement.

Allan Collard-Wexler provided us with contiguous county data. For each of the 68 state

pairs, we identify pairs of contiguous counties of adjacent states, for a total of 734 county pairs.

To further mitigate the heterogeneity of the treatment and control samples, and because proxies

for competition nationwide or over other broad geographical areas do not capture local market

competition (Ryan 2012b), we include only banks with branches in one county (“one-county

banks”) in these samples. We identify one-county banks using data on branch addresses from

the annual Summary of Deposits (SOD) surveys of the Federal Deposit Insurance Corporation

(FDIC). We include all one-county banks in each county pair with the necessary data in the pre-

and post-deregulation periods in the corresponding treatment and control samples. If either

county in a pair does not have any one-county banks with the necessary data in either the pre- or

post-deregulation periods, we delete that pair for both periods. We obtain deposits by branch

from the SOD surveys.

We obtain all financial variables for the one-county banks in the samples from the

Commercial Bank Reports of Income and Condition (Call Reports). Chartered commercial

banks must file these public reports annually with bank regulators. These reports contain banks’

balance sheets, income statements (including loan loss provisions), and other information.

Approximately 25% of the one-county banks in our sample are owned by multibank holding

20  

companies; we find in untabulated analysis that eliminating these banks yields the same

inferences as the reported empirical results.

We obtain data on county characteristics from the U.S. Census Bureau. Requiring

available data for other test and control variables described below yields a final sample of 22,569

bank-year observations for our primary analyses. On average, the one-county banks in our

sample hold 56% of the deposits in their counties in 1994, indicating that these banks play a

significant role in the counties’ economies.

In the variable designations described below, we denote firms by the subscript i, counties

by the subscript c, states by the subscript s, and time by the subscript t.

3.2. Homogeneity of the Treatment and Control Samples

We match one-county banks in contiguous counties to mitigate heterogeneity of the

treatment and control samples, except regarding deregulation of interstate branching, thereby

increasing the power of our tests and the interpretability of our results. Our primary goal in this

matching is to control for unobservable (or hard-to-observe) characteristics such as future growth

opportunities, because we control for observable characteristics in the empirical models. Despite

this fact, to provide some sense for how well this matching achieves this goal, we calculate the

absolute values of differences in observable characteristics between the treatment and control

samples in the last pre-deregulation year. We compare these calculated amounts to the absolute

values of differences in observable characteristics between the treatment sample and three

alternative reasonable but not as closely geographically or size-matched control samples in that

year: (1) all one-county banks in states in the U.S. with the same restrictions on interstate

branching as the original control states (i.e., the states of the original control banks), (2) all one-

21  

state banks in the original control states, and (3) all one-state banks in states with the same

restrictions on interstate branching as the original control states.

Specifically, we compute the average for all one-county banks in each treatment and

control county of the natural logarithm of total assets (SIZE), total equity capital divided by total

assets (CAP), earnings before loan loss provisions times 1000 divided by total assets (EARNB),

loan loss provisions times 1000 divided by lagged total loans (LLP), commercial and industrial

loans divided by total loans (C&I), total loans divided by total assets (LOANS), total deposits

divided by total liabilities (DEPOSITS), and core deposits divided by total deposits (CORE).

The Appendix provides detailed definitions of these and all other variables used in this study.

We calculate the absolute values of the differences in these averages of each of these variables

for each pair of contiguous treatment and control counties. Column 1 of Table 2 reports the

means of these absolute values across all of the pairs of contiguous counties. Columns 2-4 of the

table report the same statistics calculated using the three alternative control samples.

Comparison of column 1 to the other three columns indicates that matching based on contiguous

counties produces considerably smaller differences between treatment and control groups for all

of these variables. 

3.3. Validation that Timelier Bad News Loan Loss Provisions Increase Entry

In the hypothesis development, we argue that potential entrants are more likely to enter a

local market when incumbent banks record timelier bad news loan loss provisions. We conduct

analysis to validate this argument in this section. To do this, we develop proxies for the

timeliness of bad news and good news loan loss provisions and for future entry at the county-

year level.

22  

We build the proxies for the timeliness of bad news and good news loan loss provisions

up from the bank-year level. These proxies have two components, with the first component

capturing loan loss provision timeliness and the second capturing whether the news about loan

losses is bad or good. To calculate the first component in a simple fashion that reflects

individual incumbent banks’ loan loss provisioning in specific years, we measure loan loss

provision timeliness for a bank-year as the loan loss provision divided by ten times the next-year

change in non-performing loans.11 We calculate the weighted average of this ratio for the bank-

year observations in each county-year, where the weights equal individual banks’ shares of the

total assets of the banks in the county in that year. ctTLRATIO denotes these weighted-average

ratios.

To calculate the second component, we create two zero-one indicator variables to

distinguish bad news versus good news next-year changes in non-performing loans; 1itBNEW

( 1itGNEW ) equals one if the next-year change in non-performing loans is positive (negative) for

each bank-year observation. We calculate weighted averages of 1itBNEW and 1itGNEW across

the bank-year observations in each county-year, where the weights are the same as before. We

create two zero-one indicator variables to distinguish county-years that primarily experience bad

and good news changes in non-performing loans: 1ctBNEWC ( 1ctGNEWC ) equals one if the

weighted average of 1itBNEW is greater (less) than 0.5.

The proxy for future entry by out-of-state banks in county c in year t is the change from

year t to year t+3 in the proportion of branches in the county that are owned by out-of-state banks,

denoted ΔOUTSTct+3. We use a three-year window to allow time for out-of-state banks to enter

                                                            11 We multiply the next-year change in non-performing loans by ten for the sole purpose of making the coefficients in the validation models easier to report. Similar variable scalings elsewhere in the paper serve similar purposes.

23  

the incumbents’ local markets. In untabulated analysis, we find the use of alternative windows

to measure ΔOUTST and defining entry in terms of the change in out-of-state banks’ share of

deposits in the county yield the same inferences as the reported empirical results.

To determine the relationship between the timeliness of incumbent banks’ bad news and

good news loan loss provisions and out-of-state banks’ entry, we regress ΔOUTSTct+3 on

BNEWCct+1, TLRATIOct, the interactions between BNEWCct+1 and TLRATIOct and between

GNEWCct+1 and TLRATIOct, and control variables that capture other drivers of local market entry

by out-of-state banks discussed below.

3 0 1 1 2 1 3 1ct ct ct ct ct ctOUTST BNEWC TLRATIO BNEWC TLRATIO GNEWC

4 5 6 7 8st ct ct st stRESTR NBIZ HHI BKRUPTB BKRUPTC (1)

9 ct s t ctUNEMPL .

Because the dependent variables (as well as most of the explanatory variables) are measured at

the county-year level, we must estimate equation (1) at that level. Naturally, the state-year level

control variables discussed below take the same value for all of the corresponding county-years.

We expect incumbent banks that record timelier bad news loan loss provisions to

experience increased entry, and thus the coefficient ϕ2 on TLRATIOct×BNEWCct+1 in equation (1)

to be positive. In contrast, we expect timelier good news loan loss provisions to have no effect

on entry, and thus the coefficient ϕ3 on TLRATIOct×GNEWCct+1 to be zero. We include the

uninteracted BNEWCct+1 to ensure that the sign of the news alone does not drive either ϕ2 or ϕ3.

We have no expectations for the coefficient ϕ1 on BNEWCct+1.12

Equation (1) includes the following control variables. The state-level interest branching

restriction index ( stRESTR ) controls for the level of barriers to entry through interstate branching.

                                                            12 Entry could be induced by either bad or good news, depending on the type of investor. For example, bad news might attract Warren Buffett-type value investors, while good news might attract growth investors.

24  

The county-level number of non-government establishments in thousands ( ctNBIZ ) controls for

local market size. The county-level Herfindahl-Hirschman index of deposit concentration

( ctHHI ) controls for the existing level of competition in the local market. State-level business

and consumer bankruptcy rates (BKRUPTBst and BKRUPTCst, respectively) and the county-level

unemployment rate (UNEMPLct) control for economic conditions in the local market. We expect

negative coefficients on all of the control variables except for NBIZct, on which we expect a

positive coefficient. We include state fixed effects ( s ) and year fixed effects ( t ) to capture

other local market and time-related factors. Unless indicated otherwise, we say that a coefficient

or other statistic is statistically significant when the significance level is 5% or better in a two-

tailed test.

Panels A and B of Table 3 report the descriptive statistics and correlation matrix,

respectively, for the variables in equation (1), with all variables measured at the county-year

level. The average change in the proportion of branches owned by out-of-state banks from year t

to year t+3 (ΔOUTSTct+3) is 0.057. The average timeliness ratio (TLRATIOct) is 0.006.

ΔOUTSTct+3 is significantly negatively correlated with the interstate branching restriction

index, stRESTR , consistent with deregulation increasing entry.

Table 4 reports the estimation of equation (1). With the exception of BKRUPTBst, the

coefficients on the control variables all have the expected signs. The coefficients on NBIZct and

HHIct are significant.

We find that ΔOUTSTct+3 is significantly positively associated with incumbent banks’ bad

news loan loss provisions, as evidenced by a significantly positive coefficient ϕ2 on

TLRATIOct×BNEWCct+1 (t=5.0). In contrast, we find that ΔOUTSTct+3 is not associated with

incumbent banks’ good news loan loss provisions, as evidenced by an insignificant coefficient ϕ3

25  

on TLRATIOct×GNEWCct+1. These coefficients are significantly different (t=2.2). These results

indicate that timelier recognition of bad news loan loss provisions by incumbent banks increases

entry into local markets by out-of-state banks.13

3.4. Approach for Testing Hypothesis H1

Hypothesis H1 posits that the threat of entry by out-of-state banks is inversely related to

the timeliness of incumbents’ bad news loan loss provisions (H1a), but not to the timeliness of

their good news provisions (H1b). Similar to prior research (Gambera 2000; Nichols et al. 2009;

Beatty and Liao 2011; Bhat et al. 2012; Bushman and Williams 2012a, 2012b), we measure the

timeliness of loan loss provisions, LLPit, as the strength of their association with the next-year

change in non-performing loans, 1itNPL . 14 To test both hypotheses H1a and H1b, we

distinguish increases and decreases in year-ahead non-performing loans, 1itBNEW and

1itGNEW . We incorporate the state-level interstate branching restriction index, stRESTR ,

because we expect that incumbent banks in states more subject to entry through interstate

branching record less timely bad news loan loss provisions.

Specifically, we regress the loan loss provision for year t times 1000 divided by year t-1

total loans, itLLP , on 1itBNEW , 1itNPL , stRESTR , interactions among these variables and also

among GNEWit+1 and the latter two variables, and control variables described below:

                                                            13 Somewhat relatedly, we considered the possibility that incumbent banks managed their loan loss provisions to influence the likelihood that they were M&A targets as a consequence of IBBEA. Using the Federal Reserve’s Commercial Bank database, we identified 491 bank-year observations of acquired incumbent banks in our sample (2.2% of the sample). Inconsistent with this possibility, we find that an indicator variable for these observations is insignificantly correlated with similar measures of the timeliness of good news and bad news loan loss provisions as described in this section. 14 This approach is similar in motivation to Basu’s (1997) approach to estimating timely loss recognition. It is not subject to concerns with using (endogenous) stock returns as the measure of news, however.

26  

0 1 1 2 1 3 1 1it it st it st it itLLP BNEW RESTR BNEW RESTR NPL BNEW

4 1 1 5 1 6 1 1it it st it st it itNPL BNEW RESTR GNEW RESTR NPL GNEW

7 1 1 8 9 1 10 2 11 3it it it it it itNPL GNEW NPL NPL NPL NPL (2)

12 1 13 1 14 15 16 17it it it it it stLLA LLP CAP EARNB SIZE BKRUPTB

18 19 1 1st ct s t s it t it itBKRUPTC UNEMPL NPL NPL .

We are primarily interested in the coefficient γ3 on RESTRst×ΔNPLit+1×BNEWit+1, which

hypothesis H1a predicts is positive, because banks better protected from entry through interstate

branching should record more timely loan loss provisions. In contrast, hypothesis H1b predicts

the coefficient γ6 on RESTRst×ΔNPLit+1×GNEWit+1 is zero. We include the other variables

involving 1itBNEW , GNEWit+1, 1itNPL , and stRESTR to ensure that their omission does not

drive the estimates of γ3 and γ6. Since restrictions on interstate branching are imposed at the state

level, we calculate standard errors clustering observations by state.

To ensure that these estimates reflect the timeliness of the current loan loss provision

rather than other factors, equation (2) includes control variables appearing in the prior literature

(e.g., Beatty and Liao 2011) or that are motivated by our contiguous-county research design. To

allow for serial correlation of changes in non-performing loans, we control for the similarly

defined changes in nonperforming loans in the prior three years (ΔNPLit, ΔNPLit-1, and ΔNPLit-2)

as well as the year t-3 level of nonperforming loans divided by year t-4 total loans, denoted

NPLt-3, to capture any further lags. To allow banks’ past loan loss reserving to affect their

current loan loss provisions, we control for banks’ prior loan loss reserving using the year t-1

loan loss allowance and provision divided by year t-2 total loans ( 1itLLA and 1itLLP ,

respectively). To incorporate incumbent banks’ incentives to exercise discretion over loan loss

provisions, we control for equity capital divided by total assets ( itCAP ), earnings before loan loss

provisions divided by year t-1 total assets ( itEARNB ), and the logarithm of total assets in

27  

millions of dollars ( itSIZE ).15 While the economic conditions in contiguous counties likely are

similar, to capture any differences we control for the state-level business and consumer

bankruptcy rates (BKRUPTBst and BKRUPTCst, respectively) and the county-level

unemployment rate (UNEMPLct).

Equation (2) also includes state fixed effects, d , and year fixed effects, t , to control for

differences in the level of loan loss provisions across states and time, respectively.16 Finally, the

equation includes interactions of ΔNPLit+1 with the state (time) fixed effects to control for

differences of loan loss provision timeliness across states (time).

The research design we use to test hypothesis H1—which constructs matched treatment

and control groups of banks in contiguous counties of adjacent states and includes the control

variables and fixed effects described above in equation (2)—controls for both observable and,

more importantly, unobservable factors affecting interstate branching deregulation and/or

incumbent banks’ loan loss provision timeliness. Specifically, the matched groups and fixed

effects control for unobservable location- and time-related factors. To illustrate, one such

unobservable factor is expected future growth in the state or local economies. States anticipating

growth are more likely to deregulate banking (Kroszner and Strahan 1999). Banks may delay

recognition of expected losses to attract external financing for their investment opportunities.

Contiguous counties should be similar in their expected future growth and other unobservable

factors; to the extent they are not, the state fixed effects capture time-invariant location-related

differences.

                                                            15 Moyer (1990), Beatty et al. (1995), Collins et al. (1995), Kim and Kross (1998), Ahmed et al. (1999), and Liu and Ryan (2006) and others provide evidence that banks manage one or both of capital and earnings using loan loss provisions. 16 Replacing the state fixed effects with bank fixed effects yields similar results.

28  

Our approach is substantially similar to the differences-in-differences research design

used extensively in research in accounting and other social sciences.17 See Gormley et al. (2012,

pp. 173-174) for extensive discussion of this point in the context of an analogous research

setting—how entry by foreign banks into India affects borrowers’ timely loss recognition—and

research design.

3.4. Approach for Testing Hypothesis H2

Hypothesis H2 posits that the strength of the inverse relation between the threat of entry

by out-of-state banks and the timeliness of incumbents banks’ bad news loan loss provisions

increases with the incumbents’ proportion of heterogeneous loans (H2a) and decreases with

borrower turnover in the local loan origination market (H2b). To test hypothesis H2a (H2b), we

partition the full sample into above- and below-median heterogeneous loans (borrower turnover)

subsamples. Specifically, we compute the average of commercial and industrial loans divided by

total loans for the one-county banks in each contiguous county pair across the sample period and

partition the country pairs into above- and below-median subsamples. Similarly, we compute the

average growth of the number of non-government establishments in each continuous county pair

across the sample period and partition the county pairs into above- and below-median

subsamples.18 We expect a more positive coefficient 3 on RESTRst×ΔNPLit+1×BNEWit+1 for

the subsamples of contiguous county pairs with an above-median proportion of heterogeneous

loans or below-median borrower turnover than for the other subsamples. We expect a zero

coefficient γ6 on RESTRst×ΔNPLit+1×GNEWit+1 in all subsamples.

                                                            17 Examples of studies employing the difference-in-difference research design include Jayaratne and Strahan (1996, 1998), Bertrand and Mullainathan (2003), Dick (2006), Zarutskie (2006), Beck et al. (2010), Giroud and Mueller (2010), Callen et al. (2010), Koetter et al. (2012), and Armstrong et al. (2012). 18 Alternatively, in untabulated analyses we partitioned based on growth in population or employment in non-government establishments, both of which yield the same inferences as the reported results.

29  

4. Empirical Results

4.1. Descriptive Statistics and Correlations

Panels A and B of Table 5 report descriptive statistics and correlations, respectively, for

the variables in equation (2) for the overall (i.e., combined treatment and control) sample; due to

the large number of interactive variables in the equation, we only include statistics for the

uninteracted variables. The average total assets of banks in our sample is $74 million, consistent

with one-county banks being small. The mean (median) of earnings before loan loss provisions

times 1000 divided by total assets (EARNBit) is 13.757 (12.425); this low return on assets reflects

banks’ high financial leverage, which yields normal return on equity. The average branching

restriction index is 3.035, indicating that the average bank in the sample is exposed to over three

of the four interstate branching restrictions.

The correlations of LLPit with the explanatory variables in equation (2) generally are

consistent with those reported in prior studies. For example, LLPit is significantly positively

correlated with EARNBit, consistent with income smoothing, and also with CAPit, consistent with

more solvent banks reserving more for loan losses. LLPit is significantly positively correlated

with the current and two prior year changes in non-performing loans (ΔNPLit, ΔNPLit-1, ΔNPLit-2),

NPLit-3, and the two bankruptcy variables, consistent with banks reserving more for loan losses

when economic loan losses are higher. Perhaps surprisingly, however, LLPit is significantly

negatively correlated with the year-ahead change in non-performing loans (ΔNPLit+1). This

could be attributable to LLPit’s high correlation with net loan charge-offs, since charge-offs

reduce non-performing loans.  

30  

4.2. Regression Analyses

Table 6 reports the OLS estimation of equation (2). Hypothesis H1a posits that

incumbent banks located in states with greater restrictions on interstate branching record timelier

bad news loan loss provisions, while hypothesis H1b posits that these banks do not record

differentially timely good news loan loss provisions. Consistent with both hypotheses, the

coefficient on 1 1st it itRESTR NPL BNEW is significantly positive (t=2.2), while the

coefficient on 1 1st it itRESTR NPL GNEW is insignificant. These coefficients are not quite

significantly different, however (t=1.6). The coefficients on the control variables generally are

as expected. For example, as in prior research (Wahlen 1994; Ahmed et al. 1999; Kanagaretnam

et al. 2004; Bushman and Williams 2012a; Kilic et al. 2012), we find a significant positive

coefficient on itEARNB , consistent with the sample banks smoothing their income using loan

loss provisions.

Table 7 reports the OLS estimation of equation (2) partitioning the overall sample into

above- and below-median subsamples based on incumbent banks’ proportion of commercial and

industrial loans. Hypothesis H2a posits that incumbents that write more heterogeneous loans

record less timely bad news loan loss provisions to reduce entry by out-of-state banks.

Consistent with this hypothesis, the coefficient on 1 1st it itRESTR NPL BNEW is significantly

positive in the high commercial and industrial loans subsample (t=3.8) and insignificant in the

other subsample. These two coefficients are significantly different (t=3.6).

Interestingly, the coefficients on both ΔNPLit+1×BNEWit+1 and ΔNPLit+1×GNEWit+1 are

significantly positive in the low commercial and industrial loans subsample. These coefficients

likely reflect the fact that provisions for homogeneous loans, which rise as these loans become

31  

delinquent but are not yet nonperforming, predict near-term changes in non-performing loans

more strongly than do provisions for heterogeneous loans, as explained by Ryan (2012a).

Table 8 reports the OLS estimation of equation (2) partitioning the overall sample into

above- and below-median subsamples based on a proxy for borrower turnover, growth in the

number of non-government enterprises, in incumbent banks’ counties. Hypothesis H2b posits

that incumbent banks with lower borrower turnover record less timely bad news loan loss

provisions. Consistent with this hypothesis, the coefficient on 1 1st it itRESTR NPL BNEW is

significantly positive (t=3.1) in the low borrower turnover subsample and insignificant in the

other subsample. These two coefficients are significantly different (t=2.0).

In summary, the results of estimating equation (2) uniformly support hypotheses H1 and

H2.

5. Robustness Analyses

We conducted a number of robustness analyses for equation (2) in addition to the results

reported in Tables 6-8. To conserve space, we tabulate only the standard difference-in-

difference analyses discussed in Section 5.6. Our inferences are unaffected by all of these

analyses.

5.1. An Alternative Measure of Economic Loan Losses

In our primary analyses, we use the next-year change in non-performing loans to proxy

for economic loan losses. Alternatively, we proxy for economic loan losses by the sum of the

next-year change in non-performing loans and next-year net loan charge-offs. This proxy

captures the fact that loan charge-offs reduce non-performing loans, all else being equal.

32  

5.2. Overall Health of Incumbent Banks

A possible alternative explanation for our results is that incumbent banks record lower

loan loss provisions in an attempt to appear overall healthier (more solvent and profitable), rather

than as better loan originators. We include 1it itCAP NPL and 1it itEARNB NPL in equation

(2) to rule out this alternative explanation.

5.3. Interaction between the Proportion of Heterogeneous Loans and Borrower Turnover

We partition the sample into four subsamples based on the intersection of above- versus

below-median splits for both incumbent banks’ proportion of heterogeneous loans and borrower

turnover in incumbent banks’ counties. We estimate equation (2) for these four subsamples, and

find that the coefficient on 1 1st it itRESTR NPL BNEW is significantly positive only in the

subsample with both high proportion of heterogeneous loans and low borrower turnover. This

indicates that incumbent banks can deter entry only when they have substantial information

advantage regarding their existing borrowers and new borrowers arise relatively slowly.

5.4. Additional Drivers of the Interstate Branching Deregulation

Entry barriers arise in part from political processes shaped by interest group lobbying and

ideologies (Economides et al. 1996; Kroszner and Strahan 1999; Rice and Strahan 2010). This

raises the concern that a state’s (de)regulation of interstate branching may be correlated with

variation in the incentives of banks in the state to record timely loan loss provisions. We

performed three tests to rule out this possibility.

33  

First, in our main analyses, we include state fixed effects in equation (2) to capture

persistent factors in a state, such as industry composition. Second, following Kroszner and

Strahan (1999), we calculate six variables related to the deregulation of intrastate branch

banking, which should be shaped by related political processes in the year before the passage of

IBBEA (1993): (1) the size of the insurance sector relative to banking, (2) small bank share of all

banking assets in a state, (3) the relative capital-to-asset ratios of small versus large banks, (4)

the fraction of small nonfinancial firms, (5) whether a state is controlled by Democrats, and (6)

the share of state legislature that is Democrat. We correlate these six variables with our

interstate branching deregulation index, RESTR. Consistent with Rice and Strahan (2010), we

find that only the small bank share of all banking assets in a state and is significantly correlated

with RESTR. Third, we include six these variables as well as their interactions with ΔNPLit+1 in

equation (2).19

5.5. Additional Controls

We considered various bank and county characteristics that might affect the timeliness of

incumbent banks’ loan loss provisions, both separately and interacted with ΔNPLit+1, in equation

(2). The bank characteristics are a zero-one indicator variable that equals one if a bank’s

financial report is audited, a zero-one indicator variable that equals one if the bank is controlled

by a multi-bank holding company, and the bank’s proportions of commercial and industrial,

agriculture, consumer, and real estate loans. The county characteristics are a zero-one indicator

variable that equals one if a county belongs to a Metropolitan Statistical Area and the murder rate

                                                            19 We exclude the state fixed effects and state fixed effects interacted with ΔNPLit+1 in this specification of equation (2).

34  

in a county. The murder rate likely is related to credit risk because violent crime diminishes

economic prosperity (Henderson 1999).

5.6. Standard Difference-in-Difference Design

As discussed in Section 3.3, our research design is very similar to a difference-in-

difference design. We performed a more typically structured difference-in-difference design to

corroborate our primary results. We first sort the overall sample observations into four

subsamples: the control group in the pre-deregulation period (4,129 observations), the control

group in the post-deregulation period (7,678 observations), the treatment group in the pre-

deregulation period (3,949 observations), and the treatment group in the post-deregulation period

(6,813 observations). We then estimate an equation similar to equation (2) but without the

inclusion of interactions with RESTR, fixed effects, and other features discussed in Section 3.3

for the four subsamples. We report the coefficients on ΔNPLit+1×BNEWit+1 for the four

subsamples as well as the salient differences in these coefficients and significance levels in Table

9. We find that the coefficient on ΔNPLit+1×BNEWit+1 is significantly positive and insignificantly

different for the control and treatment groups in the pre-regulation period. From the pre- to the

post-regulation periods, the coefficient ΔNPLit+1×BNEWit+1 drops significantly for the treatment

group (t=2.9) but does not change significantly for the control group, yielding a significant

difference in the coefficient across the two groups in the post-deregulation period (t=2.0).

5.7. Clustering by Shared Borders of Adjacent States

In the primary analyses, we calculate standard errors clustering bank-year observations

by state, which adjusts for the correlation of these observations due to interstate branching

35  

deregulation or other common events. Banks in contiguous counties of adjacent states also likely

experience common events. To capture this source of cross-correlation, we instead calculate

standard errors clustering bank-year observations by shared border of adjacent states.

5.8. Weighted Least Squares

We estimated equation (2) using weighted least squares, where the weights are the

number of banks in each county-year.

6. Conclusion

While the specific benefits of timelier loan loss provisioning to banks and the financial

system as a whole are relatively well understood, the specific costs to banks are not. We argue

that potential entrants to local loan origination markets use the information in incumbent banks’

loan loss provisions to assess the incumbents’ loan underwriting quality and thus the desirability

of market entry. We further argue that, to forestall entry, the incumbents adjust the timeliness of

their loan loss provisions. We exploit an exogenous increase in the threat of entry arising from

the removal of state-level restrictions on interstate branching in the U.S. during 1994–2005 to

conduct a natural experiment. We employ a research design that compares one-county banks in

contiguous counties of adjacent states with strictly different interest branching deregulation.

We first conduct a validation analysis which shows that counties with incumbent banks

that record timelier bad news (but not good news) loan loss provisions experience more entry by

out-of-state banks in the following three years. We then show that, consistent with our

hypotheses, incumbent banks operating in states that eliminate restrictions on interstate

branching delay recognition of bad news (but not good news) loan loss provisions, and that this

36  

discretionary behavior is concentrated among banks in counties with a high proportion of

heterogeneous loans and a low degree of borrower turnover.

Our evidence suggests that credit competition plays a significant role in bank’s

discretionary loan loss provisioning. Accordingly, extant proposals to require timelier provisions

may entail unintended consequences for the competitiveness of individual banks.

37  

References

Ahmed, A., C. Takeda, and S. Thomas. 1999. Bank loan loss provisions: A reexamination of

capital management, earnings management and signaling effects. Journal of Accounting and Economics 28: 1-25.

Armstrong, C., K. Balakrishnan, and D. Cohen. 2012. Corporate governance and the information

environment: Evidence from state antitakeover laws. Journal of Accounting and Economics 53: 185-204.

Ashcraft, A. 2005. Are banks really special? New evidence from the FDIC-induced failure of

healthy banks. American Economic Review 95: 1712-1730. Basu, S. 1997. The conservatism principle and the asymmetric timeliness of earnings. Journal of

Accounting and Economics 24: 3–37. Beatty, A., S. Chamberlain, and J. Magliolo. 1995. Managing financial reports of commercial

banks: The influence of taxes, regulatory capital, and earnings. Journal of Accounting Research 33: 231-261.

Beatty, A., and S. Liao. 2011. Do delays in expected loss recognition affect banks’ willingness to

lend? Journal of Accounting and Economics 52: 1-20. Beck, T., R. Levine, and A. Levkov. 2010. Big Bad Banks? The winners and losers from bank

deregulation in the United States. Journal of Finance 65: 1637-1667. Bernanke, B. 2009. Lessons of the financial crisis for banking supervision. Speech at the Federal

Reserve Bank of Chicago Conference on Bank Structure and Competition. Berger, A., R. Demsetz, and P. Strahan. 1999. The consolidation of the financial services

industry: Causes, consequences, and implications for the future. Journal of Banking and Finance 23: 135-194.

Berger, A., and A. Dick. 2007. Entry in banking markets and the early-mover advantage. Journal

of Money, Credit and Banking 39: 775-807. Berger, A, A. Kashyap, and J. Scalise. 1995. The transformation of the U.S. banking industry:

What a long, strange trip it’s been. Brooking Papers on Economic Activity 2: 55-218. Bertrand, M. and S. Mullainathan. 2003. Enjoying the quite life? Managerial behavior following

anti-takeover legislation. Journal of Political Economy 111: 1042-1075. Bhat, G., S. Ryan, and D. Vyas. 2012. The implications of banks’ credit risk modeling for their

loan loss provision timeliness and loan origination procyclicality. Working Paper. New York University.

38  

Black, S. 1999. Do better schools matter? Parental valuation of elementary education. Quarterly

Journal of Economics 114: 577-599. Burks, J., C. Cuny, J. Gerakos, and J. Granja. 2012. Entry Threats and Voluntary Disclosure.

Working Paper. University of Chicago. Bushman, R., and C. Williams. 2012a. Accounting discretion, loan loss provisioning, and

discipline of banks’ risk-taking. Journal of Accounting and Economics 54: 1-18. Bushman, R., and C. Williams. 2012b. Delayed expected loss recognition and the risk profile of

banks. Working Paper. University of North Carolina-Chapel Hill. Bushman, R., B. Hendricks, and C. Williams. 2013. Perceived bank competition: Operational

Decision-making and bank stability. Working paper, University of North Carolina-Chapel Hill.

Callen, J., Y. Guan, and J. Qiu. 2010. The market for corporate control and accounting

conservatism. Working Paper. University of Toronto. Calomiris, C., J. Mason. 2003. Consequences of bank distress during the Great Depression.

American Economic Review 93: 937-947. Card, D., and A. Kruger. 1994. Minimum wages and employment: A case study of the fast-food

industry in New Jersey and Pennsylvania. American Economic Review 84: 772-793. Clair, R., G. O’Driscoll Jr., and K. Yeats. 1994. Is banking different? A re-examination of the

case for regulation. CATO Journal 13: 345-358. Collins, J., D. Shackelford, and J. Wahlen. 1995. Bank differences in the coordination of

regulatory capital, earnings, and taxes. Journal of Accounting Research 33: 263-291. Darrough, M. 1993. Disclosure policy and competition: Cournot vs. Bertrand. The Accounting

Review 68 (3): 534-561. Darrough, M. and N. Stoughton. 1990, Financial disclosure policy in an entry game, Journal of

Accounting and Economics 12: 219-243. Dell’Ariccia, G., E. Friedman, and R. Marquez. 1999. Adverse selection as a barrier to entry in

the banking industry. Rand Journal of Economics 30: 515-534. Dell’Ariccia, G., and R. Marquez. 2004. Information and bank credit allocation. Journal of

Financial Economics 72: 185-214. Dell’Ariccia, G., and R. Marquez. 2006. Lending booms and lending standards. Journal of

Finance 61: 2511-2546.

39  

Dell’Ariccia, G., and R. Marquez. 2008. Can cost increases increase competition? Asymmetric

information and equilibrium prices. RAND Journal of Economics 39: 144-162. Dick, A. 2006. Nationwide branching and its impact on market structure, quality, and bank

performance. Journal of Business 79: 567-592. Docking, D., M. Hirschey, and E. Jones. 1997. Information and contagion effects of bank loan-

loss reserve announcements. Journal of Financial Economics 43: 219-239. Dugan, J. 2009. Remarks before the institute of international bankers, loan loss provisioning and

pro-cyclicality. Economides, N., G. Hubbard, and D. Palia. 1996. The political economy of branching

restrictions and deposit insurance: A model of monopolistic competition among small and large banks. Journal of Law and Economics: 39: 667-704.

Financial Stability Forum. 2009. Report of the financial stability forum on addressing

procyclicality in the financial system. Fox, W. 1986. Tax structure and the location of economic activity along state borders. National

Tax Journal 39: 387-401. Gambera, M., 2000. Simple forecasts of bank loan quality in the business cycle. Working Paper.

Federal Reserve Bank of Chicago. Gan, J., and T. Riddiough. 2008. Monopoly and information advantage in the residential

mortgage market. Review of Financial Studies 21: 2677-2703. Garmaise, M., and T. Moskowitz. 2006. Bank mergers and crime: The real and social effects of

credit market competition. Journal of Finance 61: 495-538. Gilbert, R., and L. Kochin. 1989. Local economic effects of bank failures. Journal of Financial

Services Research 3: 333-345. Giroud, X., and H. Mueller. 2010. Does corporate governance matter in competitive industries?

Journal of Financial Economics, 95: 312-331. Golembe, G. 1994. History offers some clues on significance of interstate branching. Banking

Policy Report 13: 4-4. Gormley, T., B. Kim, and X. Martin. 2012. Do firms adjust their timely loss recognition in

response to changes in the banking industry? Journal of Accounting Research 50: 159-196.

40  

Henderson, C. 1999. The economic performance of African-American-Owned Banks: The role of loan loss provisions. American Economic Review Papers and Proceedings: 89: 372-376.

Holmes, T. 1998. The effect of state policies on the location of manufacturing: Evidence from

state borders. Journal of Political Economy 106: 667-705. Huang, R. 2008. Evaluating the real effect of bank branching deregulation: Comparing

contiguous counties across US state borders. Journal of Financial Economics 87: 678-705.

Huber, G., and K. Arceneaux. 2007. Identifying the persuasive effects of presidential advertising.

American Journal of Political Science 51: 957-977. Hutnyan, J. 1995. States start considering whether or when to opt for interstate branching.

Banking Policy Report 14: 7-7. Jayaratne, J., and P. Strahan. 1996. The finance-growth nexus: Evidence from bank branch

deregulation. Quarterly Journal of Economics 111: 639-670. Jayaratne, J., and P. Strahan. 1998. Entry restrictions, industry evolution and dynamic efficiency:

Evidence from commercial banking. Journal of Law and Economics 41: 239-274. Johnson, C., and T. Rice. 2008. Assessing a decade of interstate bank branching. The

Washington and Lee Law Review 65: 73-127. Kanagaretnam, K., G. Lobo, and D. Yang. 2004. Joint tests of signaling and income smoothing

through bank loan loss provisions. Contemporary Accounting Research 21: 843-884. Keele, L., and R. Titiunik. 2012. Geography as a causal variable. Working Paper. University of

Michigan. Kilic, E, G. Lobo, T. Ranasinghe, and S. Sivaramakrishnan. 2012. The impact of SFAS 133 on

income smoothing by banks through loan loss provisions. The Accounting Review forthcoming.

Kim, M, and W. Kross. 1998. The impact of the 1989 change in bank capital standards on loan

loss provisions and loan write-offs. Journal of Accounting and Economics 25: 69-99. Koetter, M, J. Kolari and L. Spierdijk. 2012. Enjoying the quiet life under deregulation?

Evidence from adjusted Lerner indices from U.S. banks. Review of Economics and Statistics 94: 462-480.

Kothari, S.P., K. Ramanna, and D. Skinner. 2010. Implications for GAAP from an analysis of

positive research in accounting. Journal of Accounting and Economics 50: 246-286.

41  

Kroszner, R., and P. Strahan. 1999. What drives deregulation? Economics and politics of the relaxation of bank branching restrictions. Quarterly Journal of Economics 114: 1437-1467.

Liu, C., and S. Ryan. 1995. The effect of bank loan portfolio composition on the market reaction

to and anticipation of loan loss provisions. Journal of Accounting Research 33: 77-94. Liu, C., and S. Ryan. 2006. Income smoothing over the business cycle: Changes in banks’

coordinated management of provisions for loan losses and loan charge-offs from the pre-1990 bust to the 1990s boom. The Accounting Review 81: 421-441.

Liu, C., S. Ryan, and J. Wahlen. 1997. Differential valuation implications of loan loss provisions

across banks and fiscal quarters. The Accounting Review 72: 133-146. Marquez, R. 2002. Competition, adverse selection, and information dispersion in the banking

industry. Review of Financial Studies 15: 901-926. Mester, L., L. Nakamura, M. Renault. 2007. Transactions accounts and loan monitoring. Review

of Financial Studies 20: 529-556. Milgrom, P., and J. Roberts. 1982. Limit pricing and entry under incomplete information: An

equilibrium analysis. Econometrica 50: 443-459. Moyer, S. 1990. Capital Adequacy ratio regulations and accounting choices in commercial banks.

Journal of Accounting and Economics 13: 123-154. Nichols, D., J. Wahlen, and M. Wieland. 2009. Publicly traded versus privately held:

Implications for conditional conservatism in bank accounting. Review of Accounting Studies 14: 88-122.

Pagano and Jappelli. 1993. Information sharing in credit markets. Journal of Finance 43: 1693-

1718. Petersen, M., and R. Rajan. 2002. Does distance still matter? The information revolution in small

business lending. Journal of Finance 57: 2533-2570. Prager, R., and T. Hannan. 1998. Do substantial horizontal mergers generate significant price

effects? Evidence from the banking industry. Journal of Industrial Economics 46: 433-452.

Rajan, R. 1992. Insiders and outsiders, the choice between informed and arm’s-length debt.

Journal of Finance 47: 1367-1400. Redlich, F. 1951. The Molding of American Banking: Men and Ideas, Part II, 1840-1920. New

York: Hafner Publishing Company.

42  

Rhoades, S. 2000. Bank mergers and banking structure in the US, 1980-1998. Board of Governors Staff Study 174. Board of Governors of the Federal Reserve System, Washington.

Rice, T., and P. Strahan. 2010. Does credit competition affect small-firm finance? Journal of

Finance 65: 861-889. Ryan, S. G. 2007. Financial instruments and institutions: Accounting and disclosure rules

(Second Edition). New Jersey: John Wiley & Sons. Ryan, S. G. 2012a. Financial reporting for financial instruments. Foundations and Trends in

Accounting 6: 3-4, pp. 187-354. Ryan, S. P. 2012b. The costs of environmental regulation in a concentrated industry.

Econometrica 80: 1019-1061. Sylla, R., J. Legler, and J. Wallis. 1987. Banks and state public finance in the new republic: The

United States, 1790-1860. Journal of Economic History 47: 391-403. Strahan, P. 2003. The real effects of U.S. banking deregulation. Federal Reserve Bank of St.

Louis Review 85: 111-128. Taylor, A., and A. Sansone. 2007. The Handbook of Loan Syndications and Trading. McGraw-

Hill. U.S. Treasury Report. 2009. A new foundation: Rebuilding financial supervision and regulation

Financial Regulatory Reform. Verrecchia, R. 1990. Endogenous proprietary costs through firm interdependence. Journal of

Accounting and Economics 12: 245-250. Wahlen, J. 1994. The nature of information in commercial bank loan loss disclosures. The

Accounting Review 69: 455-478. Zarutskie, R. 2006. Evidence on the effects of bank competition on firm borrowing and

investment. Journal of Financial Economics 81: 503-537. Zou, Y. 2012. Strategic entry decisions, accounting signals, and risk management disclosure.

Working Paper. University of Toronto.

43  

APPENDIX Variable Definitions

Bank variables: LLP Provision for loan and lease losses (RIAD4230) multiplied by 1000 and

divided by lagged total loans (RCFD1400). NPL The sum of total loans and lease finance receivables nonaccrual

(RCFD1403) and total loans and lease finance receivables past due 90 days or more and still accruing (RCFD1407) divided by lagged total loans (RCFD1400). 

∆NPL Change of NPL.

BNEW An indicator variable that takes the value of one if ∆NPL ≥ 0.

GNEW An indicator variable that takes the value of one if ∆NPL < 0.

LLA Allowance for loan and lease losses (RCFD3123) divided by lagged total loans (RCFD1400).

SIZE The logarithm of total assets (RCFD2170) in millions of U.S. dollars.

CAP Total equity capital (RCFD3210) divided by total assets (RCFD2170).

EARNB The sum of income before extraordinary items and other adjustments (RIAD4300) and provision for loan and lease losses (RIAD4230). multiplied by 1000 and divided by total assets (RCFD2170).

C&I Commercial and industrial loans (RCFD1600) divided by total loans (RCFD1400)

LOANS Total loans (RCFD1400) divided by total assets (RCFD2170). DEPOSITS Total deposits (RCFD2200) divided by total liabilities, which are calculated

as total assets (RCFD2170) minus total equity capital (RCFD3210). CORE Core deposits divided by total deposits (RCFD2200). Core deposits are

total deposits minus the aggregate balance of each deposit account with a balance of more than $100,000 (RCON2710).

State variables: MINAGE An indicator variable that takes the value of one if a state imposes a

minimum age of three years or more on target institutions.

DEPCAP An indicator variable that takes the value of one if a state imposes a statewide cap of less than 30% on deposits that any one bank could hold.

DENOVO An indicator variable that takes the value of one if a state does not permit out-of-state banks to open new branches.

SGLBR An indicator variable that takes the value of one if a state disallows out-of-state banks from acquiring single branches or portions of a banking institution.

RESTR The sum of MINAGE, DEPCAP, DENOVO, and SGLBR.

44  

BKRUPTB The number of business bankruptcy filings divided by the number of non-government establishments.

BKRUPTC The number of consumer bankruptcy filings divided by the number of population.

 

County variables: TLRATIO Timeliness ratio at the county-year level, calculated as the average of the

timeliness ratio of all one-county banks in that county for each year, weighted by each bank’s total assets. A bank’s timeliness ratio is loan loss provisions divided by next year’s change in non-performing loans, divided by 10.

BNEWC An indicator variable that takes the value of one if the mean of BNEW of all one-county banks in that county for each year, weighted by each bank’s total assets, is greater than or equal to 0.5.

GNEWC An indicator variable that takes the value of one if the mean of BNEW of all one-county banks in that county for each year, weighted by each bank’s total assets, is less than 0.5.

ΔOUTST The change in the proportion of branches owned by out-of-state banks in a county.

NBIZ The number of non-government establishment in a county, divided by 1000.

HHI Herfindahl-Hirschman Index of deposit concentration among banks in a county.

UNEMPL Unemployment rate.

Note: Call Report variable designations are in parentheses

45  

Figure 1 Contiguous Counties of Neighboring States with

Strictly Different Interstate Branching Deregulation

Deregulated County

Contiguous County with Tighter Restrictions

46  

Table 1 Adjacent States Exhibiting Strictly Different Interstate Branching Deregulation

Less Restr- More Restr- Less Restr- More Restr-

Restricted iction Restricted iction Restricted iction Restricted iction Border Period State Index State Index Border Period State Index State Index

1 1997-2005 Alabama 3 Mississippi 4 38 1997-1999 West Virginia 1 Kentucky 4 2 1998-2001 Tennessee 2 Alabama 3 2000-2005 West Virginia 1 Kentucky 3

2002-2005 Tennessee 1 Alabama 3 39 1997-2005 Louisiana 3 Mississippi 4 3 1997-2001 Arizona 3 Colorado 4 40 1997-1999 Louisiana 3 Texas 4

2002-2005 Arizona 2 Colorado 4 41 1997-2000 Maine 0 New Hampshire 4 4 1995-2001 Utah 2 Arizona 4 2001 Maine 0 New Hampshire 1

1997-2005 Utah 2 Arizona 3 42 1996 Maryland 0 West Virginia 4 5 1997-2005 Louisiana 3 Arkansas 4 1997-2005 Maryland 0 West Virginia 1 6 2000-2005 Oklahoma 1 Arkansas 4 43 1997-2000 Massachusetts 1 New Hampshire 4 7 1997 Tennessee 3 Arkansas 4 44 1997-2005 Massachusetts 1 New York 2

1998-2001 Tennessee 2 Arkansas 4 45 1995-1996 Rhode Island 0 Massachusetts 4 2002-2005 Tennessee 1 Arkansas 4 1997-2005 Rhode Island 0 Massachusetts 1

8 2000-2005 Texas 1 Arkansas 4 46 1997 Tennessee 3 Mississippi 4 9 1996-2005 New Mexico 3 Colorado 4 1998-2001 Tennessee 2 Mississippi 4 10 2000-2005 Oklahoma 1 Colorado 4 2002-2005 Tennessee 1 Mississippi 4 11 1995-2000 Utah 2 Colorado 4 47 2000-2005 Oklahoma 1 Missouri 4

2001-2005 Utah 1 Colorado 4 48 1997 Tennessee 3 Missouri 4 12 1997-2005 Wyoming 3 Colorado 4 1998-2001 Tennessee 2 Missouri 4 13 1995-1996 Connecticut 1 New York 4 2002-2005 Tennessee 1 Missouri 4

1997-2005 Connecticut 1 New York 2 49 1997-2003 North Dakota 3 Montana 4 14 1995-2005 Rhode Island 0 Connecticut 1 2004-2005 North Dakota 1 Montana 4 15 1996-2005 Maryland 0 Delaware 3 50 1996-2005 South Dakota 3 Montana 4 16 1996-2005 New Jersey 1 Delaware 3 51 1997-2005 Wyoming 3 Montana 4 17 1996-2005 Pennsylvania 0 Delaware 3 52 1996-2005 South Dakota 3 Nebraska 4 18 1996 North Carolina 0 Georgia 4 53 1997-2005 Wyoming 3 Nebraska 4

1997-2005 North Carolina 0 Georgia 3 54 1995-2005 Utah 2 Nevada 4 19 1998-2001 Tennessee 2 Georgia 3 1996-2000 Utah 2 Nevada 3

2002-2005 Tennessee 1 Georgia 3 55 1996-2000 Vermont 2 New Hampshire 4 20 1996-2005 Nevada 3 Idaho 4 2001 Vermont 0 New Hampshire 1 21 1998-2005 Oregon 3 Idaho 4 56 1996 New Jersey 1 New York 4 22 1995-2000 Utah 2 Idaho 4 1997-2005 New Jersey 1 New York 2

2001-2005 Utah 1 Idaho 4 57 1996-2005 Pennsylvania 0 New Jersey 1 23 1996-2004 Washington 3 Idaho 4 58 1996-1999 New Mexico 3 Oklahoma 4

2005 Washington 1 Idaho 4 59 1996-1999 New Mexico 3 Texas 4 24 1997-2005 Wyoming 3 Idaho 4 60 1995-2005 Utah 2 New Mexico 4 25 1997 Indiana 0 Illinois 4 1996-2000 Utah 2 New Mexico 3

1998 Indiana 0 Illinois 3 61 1996 Pennsylvania 0 New York 4 1999-2004 Indiana 1 Illinois 3 1997-2005 Pennsylvania 0 New York 2

26 1998-2004 Illinois 3 Iowa 4 62 1996 North Carolina 0 South Carolina 4 2005 Illinois 0 Iowa 4 1997-2005 North Carolina 0 South Carolina 3

27 1998-2004 Illinois 3 Missouri 4 63 1996 North Carolina 0 Tennessee 4 2005 Illinois 0 Missouri 4 1997 North Carolina 0 Tennessee 3

28 1997-1998 Indiana 0 Kentucky 4 1998-2001 North Carolina 0 Tennessee 2 1999 Indiana 1 Kentucky 4 2002-2005 North Carolina 0 Tennessee 1

29 1999-2005 Ohio 0 Indiana 1 64 1997-2005 Ohio 0 West Virginia 1 30 1997-2005 Minnesota 3 Iowa 4 65 1996 Pennsylvania 0 West Virginia 4 31 1996-2005 South Dakota 3 Iowa 4 1997-2005 Pennsylvania 0 West Virginia 1 32 1996-2005 Wisconsin 3 Iowa 4 66 1996 Virginia 0 Tennessee 4 33 2000-2005 Oklahoma 1 Kansas 4 1997 Virginia 0 Tennessee 3 34 2000-2005 Kentucky 3 Missouri 4 1998-2001 Virginia 0 Tennessee 2 35 1997-1999 Ohio 0 Kentucky 4 2002-2005 Virginia 0 Tennessee 1

2000-2005 Ohio 0 Kentucky 3 67 1995-2005 Utah 2 Wyoming 4 36 1997 Tennessee 3 Kentucky 4 1997-2000 Utah 2 Wyoming 3

1998-1999 Tennessee 2 Kentucky 4 68 1996 Virginia 0 West Virginia 4 37 1996-1999 Virginia 0 Kentucky 4 1997-2005 Virginia 0 West Virginia 1

2000-2005 Virginia 0 Kentucky 3

47  

Table 2 Benefits of Using Matched Treatment and Control Samples of One-County Banks from Contiguous Counties of

Neighboring States with Strictly Different Interstate Branching Deregulation

Absolute Value of the Difference Between One-County Banks in Treatment Sample and One-County Banks One-State Banks

Original Control Sample of in any State with One-State Banks in any State with One-County Banks Same Regulation as in Original Same Regulation as

in Contiguous Counties Original Control State Control State Original Control State (1) (2) (3) (4)

Mean Mean Mean Mean SIZE 0.672 0.781 0.960 1.128 CAP 0.025 0.028 0.030 0.033 EARNB 4.239 5.011 4.754 13.154 LLP 4.451 5.353 4.344 6.411 C&I 0.063 0.077 0.080 0.098 LOANS 0.102 0.118 0.129 0.138 DEPOSITS 0.036 0.038 0.049 0.055 CORE 0.071 0.081 0.082 0.099

(5) (6) (7) (8) Median Median Median Median

SIZE 0.523 0.653 0.743 0.884 CAP 0.017 0.019 0.021 0.022 EARNB 2.442 3.045 3.043 3.651 LLP 2.016 2.261 2.466 2.538 C&I 0.045 0.060 0.058 0.074 LOANS 0.090 0.098 0.107 0.113 DEPOSITS 0.019 0.020 0.024 0.024 CORE 0.052 0.065 0.062 0.076

48  

Table 3 Descriptive Statistics for the Validation Test

Panel A: Summary Statistics for the Sample at the County-Year Level Variable Mean STD P25 Median P75 ∆OUTSTct+3 0.057 0.128 0.000 0.000 0.095 BNEWCct+1 0.490 0.500 0.000 0.000 1.000 TLRATIOct× BNEWCct+1 0.137 2.229 0.000 0.000 0.032 TLRATIOct× GNEWCct+1 -0.130 1.457 -0.048 0.000 0.000 RESTRst 2.921 1.420 2.000 4.000 4.000 NBIZct 2.812 5.636 0.331 0.685 2.132 HHIct 0.262 0.154 0.161 0.224 0.320 BKRUPTBst 0.006 0.005 0.004 0.005 0.007 BKRUPTCst 0.005 0.002 0.003 0.005 0.006 UNEMPLct 5.280 2.135 3.700 4.800 6.300 Panel B: Pearson Correlation Matrix for the Sample at the County-Year Level

∆OUTSTct+3 BNEWCct+1 TLRATIOct× BNEWCct+1

TLRATIOct× GNEWCct+1 RESTRst NBIZct HHIct BKRUPTBst BKRUPTCst

BNEWCct+1 -0.046*

TLRATIOct× BNEWCct+1 0.036* 0.063*

TLRATIOct× GNEWCct+1 0.008 0.087* 0.006

RESTRct -0.041* 0.024* 0.007 -0.011 NBIZct 0.156* -0.037* 0.002 0.008 -0.220* HHIct -0.110* 0.020* -0.017 -0.042* 0.122* -0.281* BKRUPTBst -0.023* -0.020* 0.001 -0.049* 0.048* -0.001 0.083* BKRUPTCst -0.012 0.015 0.004 -0.018 -0.317* -0.067* 0.008 -0.077* UNEMPLct 0.028* -0.020* 0.008 0.029* 0.029* -0.095* 0.105* -0.037* 0.147* Notes: N=7,435. *Denotes significance at the 10% level (using a two-tailed test).

49  

Table 4 Loan Loss Provision Timeliness and Out-of-State Bank Entry

∆OUTSTct+3 BNEWCct+1 -0.009**

(2.56)

TLRATIOct×BNEWCct+1 0.002*** (5.07)

TLRATIOct×GNEWCct+1 0.001 (1.15)

RESTRst -0.009*

(1.85)

NBIZct 0.002*** (7.31)

HHIct -0.095*** (3.56)

BKRUPTBst 0.552

(0.97)

BKRUPTCst -2.204

(0.42)

UNEMPLct -0.000

(0.09) Constant 0.091*

(1.96)

State FE Yes Year FE Yes Observations 7435 Adj. R-squared 0.133 Notes: The table reports the estimation of equation (1). The absolute values of the t statistics are calculated clustering observations by state are presented in parentheses.* (**) [***] denotes significance at the 0.10 (0.05) [0.01] level in a two-tailed test. Variables are defined in the Appendix.

50  

Table 5 Descriptive Statistics for the Main Analyses

Panel A: Summary Statistics for the Sample at the Bank-Year Level Variable Mean STD P25 Median P75LLPit 0.335 0.447 0.015 0.177 0.419RESTRst 3.035 1.301 3.000 4.000 4.000BNEWit+1 0.500 0.500 0.000 0.000 1.000∆NPLit+1 0.000 0.010 -0.004 0.000 0.003∆NPLit 0.000 0.011 -0.004 0.000 0.003∆NPLit-1 -0.001 0.011 -0.005 0.000 0.003∆NPLit-2 -0.001 0.012 -0.005 0.000 0.003NPLit-3 0.013 0.015 0.002 0.008 0.019LLPit-1 0.004 0.009 0.000 0.002 0.004LLAit-1 0.018 0.024 0.011 0.015 0.020CAPit 0.109 0.040 0.083 0.098 0.122EARNBit 0.014 0.009 0.009 0.012 0.016SIZEit 4.307 1.010 3.566 4.200 4.931BKRUPTBst 0.007 0.006 0.004 0.006 0.007BKRUPTCst 0.005 0.002 0.003 0.004 0.006UNEMPLct 5.172 1.938 3.800 4.800 6.100 Panel B: Pearson Correlation Matrix for the Sample at the Bank-Year Level LLPit RESTRst BNEWit+1 ∆NPLit+1 ∆NPLit ∆NPLit-1 ∆NPLit-2 NPLit-3 LLPit-1 LLAit-1 CAPit EARNBit SIZEit BKRUPTBst BKRUPTCst RESTRst 0.007 BNEWit+1 -0.064* 0.030* ∆NPLit+1 -0.072* 0.016* 0.664* ∆NPLit 0.079* 0.003 -0.180* -0.304*∆NPLit-1 0.087* -0.013* -0.029* -0.049* -0.278*∆NPLit-2 0.052* -0.033* -0.051* -0.052* -0.040* -0.253*NPLit-3 0.148* 0.029* -0.095* -0.075* -0.129* -0.198* -0.437*LLPit-1 0.598* -0.006 -0.044* -0.039* -0.058* 0.079* 0.066* 0.192*LLAit-1 0.152* 0.053* -0.029* -0.042* -0.044* 0.066* 0.029* 0.247* 0.310* CAPit 0.055* -0.056* 0.015* -0.008 -0.002 0.002 0.022* -0.005 0.120* 0.225*EARNBit 0.474* 0.043* 0.004 0.014* 0.002 0.010 0.006 0.030* 0.525* 0.232* 0.263*SIZEit 0.168* -0.203* -0.036* -0.023* -0.005 -0.012* -0.006 0.043* 0.212* 0.085* -0.020* 0.304*BKRUPTBst 0.203* 0.074* 0.000 -0.003 -0.003 -0.002 -0.018* 0.123* 0.235* 0.088* 0.118* 0.247* 0.087*BKRUPTCst 0.035* -0.298* 0.002 0.014* 0.027* 0.050* 0.061* -0.085* -0.022* -0.077* 0.002 -0.036* -0.018* -0.099*UNEMPLct 0.008 0.050* -0.004 -0.007 -0.007 0.000 -0.005 0.092* -0.030* -0.009 -0.036* -0.046* -0.015* -0.087* 0.170* Notes: N=22,569. *Denotes significant at the 10% level in a two-tailed test.

51  

Table 6 Interstate Branching Restrictions and Loan Loss Provision Timeliness

LLPit BNEWit+1 0.023

(0.09) RESTRst×BNEWit+1 -0.033

(0.61) RESTRst×∆NPLit+1×BNEWit+1 17.506**

(2.16) ∆NPLit+1×BNEWit+1 5.928

(0.16) RESTRst×GNEWit+1 -0.012

(0.14) RESTRst×∆NPLit+1×GNEWit+1 2.087

(0.24) ∆NPLit+1×GNEWit+1 12.397

(0.30) ∆NPLit 77.860***

(10.60) ∆NPLit-1 84.964***

(15.94) ∆NPLit-2 78.090***

(9.70) NPLit-3 69.226***

(9.01) LLPit-1 0.204***

(23.07) LLAit-1 -14.266***

(5.06) CAPit -7.862***

(9.25) EARNBit 0.115***

(14.87) SIZEit 0.034

(0.45) BKRUPTBst -4.421

(1.25) BKRUPTCst 118.542

(1.33)

52  

UNEMPLct 0.043 (1.47)

Constant -0.390 (0.50)

State FE Yes Year FE Yes State FE×∆NPLit+1 Yes Year FE×∆NPLit+1 Yes Observations 22569 Adj. R-squared 0.475 Notes: The table reports the estimation of equation (2). The absolute values of the t-statistics based on the standard errors clustered by state are presented in parentheses. * (**) [***] denotes significant at the 0.10 (0.05) [0.01] level in a two-tailed test. Variables are defined in the Appendix.

53  

Table 7 Partition by Proportion of Commercial and Industrial Loans

Proportion of C&I Loans

High Low BNEWit+1 0.236 -0.231

(0.72) (0.80) RESTRst×BNEWit+1 -0.136 0.027

(1.53) (0.42) RESTRst×∆NPLit+1×BNEWit+1 44.892*** -2.145

(3.83) (0.31) ∆NPLit+1×BNEWit+1 -57.662 283.464***

(1.29) (3.91) RESTRst×GNEWit+1 -0.016 -0.069

(0.14) (0.85) RESTRst×∆NPLit+1×GNEWit+1 15.237 -10.799

(0.97) (1.12) ∆NPLit+1×GNEWit+1 4.103 263.537***

(0.07) (3.73) ∆NPLit 76.588*** 79.899***

(13.90) (6.09) ∆NPLit-1 86.416*** 83.649***

(15.44) (9.44) ∆NPLit-2 76.731*** 80.245***

(16.20) (5.56) NPLit-3 71.399*** 66.467***

(7.78) (4.69) LLPit-1 0.203*** 0.206***

(11.23) (24.09) LLAit-1 -20.145*** -12.712***

(2.82) (5.98) CAPit -6.332*** -9.232***

(3.78) (9.06) EARNBit 0.114*** 0.117***

(11.00) (13.74) SIZEit 0.103 -0.056

(0.95) (0.73) BKRUPTBst -8.254 -2.554

(0.20) (0.74) BKRUPTCst 163.828 80.939

(1.47) (0.86)

54  

UNEMPLct 0.006 0.088** (0.18) (2.40)

Constant -0.848 0.301 (1.01) (0.37)

State FE Yes Yes Year FE Yes Yes State FE×∆NPLit+1 Yes Yes Year FE×∆NPLit+1 Yes Yes Observations 11244 11325 Adj. R-squared 0.305 0.601 Notes: The table reports the estimation of equation (2) for subsamples of county pairs for which incumbent banks have above- versus below-median proportions of commercial and industrial loans. The absolute values of the t-statistics based on the standard errors clustered by state are presented in parentheses. * (**) [***] denotes significant at the 0.10 (0.05) [0.01] level in a two-tailed test. Variables are defined in the Appendix.

55  

Table 8 Partitioning by Growth in the Number of Non-government Establishments

Growth in the Number of Non-government Establishments

Low High BNEWit+1 0.338 -0.313

(0.72) (1.36) RESTRst×BNEWit+1 -0.073 0.015

(0.78) (0.23) RESTRst×∆NPLit+1×BNEWit+1 27.837*** 2.365

(3.07) (0.19) ∆NPLit+1×BNEWit+1 56.294* 34.973

(1.98) (0.60) RESTRst×GNEWit+1 0.060 -0.090

(0.46) (1.30) RESTRst×∆NPLit+1×GNEWit+1 12.609 -12.461

(0.95) (1.46) ∆NPLit+1×GNEWit+1 65.670 38.299

(1.54) (0.81) ∆NPLit 74.677*** 81.748***

(10.57) (7.43) ∆NPLit-1 88.224*** 80.337***

(16.45) (8.79) ∆NPLit-2 85.679*** 71.278***

(11.59) (5.24) NPLit-3 80.964*** 60.378***

(15.79) (3.92) LLPit-1 0.199*** 0.210***

(15.60) (14.18) LLAit-1 -35.958*** -12.218***

(3.57) (6.73) CAPit -7.795*** -7.982***

(4.21) (9.98) EARNBit 0.124*** 0.111***

(7.28) (13.66) SIZEit 0.023 0.051

(0.19) (0.63) BKRUPTBst -16.502 -3.680

(0.49) (1.00) BKRUPTCst 147.602 143.717

(1.50) (1.20)

56  

UNEMPLct 0.025 -0.000 (0.54) (0.00)

Constant -0.801 0.133 (0.66) (0.16)

State FE Yes Yes Year FE Yes Yes State FE×∆NPLit+1 Yes Yes Year FE×∆NPLit+1 Yes Yes Observations 11293 11276 Adj. R-squared 0.329 0.594 Notes: The table reports the estimation of equation (2) for subsamples of county pairs for which the paired counties have below- versus above-median growth in the number of non-government establishments. The absolute values of the t-statistics based on the standard errors clustered by state are presented in parentheses. * (**) [***] denotes significant at the 0.10 (0.05) [0.01] level in a two-tailed test. Variables are defined in the Appendix.

57  

Table 9 Standard Difference-in-Differences Approach

control treatment control minus treatment pre-deregulation 66.876**

(2.24) [4129]

77.224***(5.82) [3949]

-10.348 (-0.32)

post-deregulation 60.587***(4.52) [7678]

20.773 (1.46) [6813]

39.814** (2.04)

pre-deregulation minus post- deregulation 6.289 (0.19)

56.451***(2.90)

Notes: The table presents selected results from the estimation of an equation similar to equation (2)—but without the inclusion of interactions with RESTR, fixed effects, and other features—for the control and treatment groups in the pre- and post-deregulation periods. Specifically, the table reports the coefficients on ΔNPLit+1×BNEWit+1, salient differences of these coefficients across the four samples, and significance levels. The absolute values of the t-statistics based on the standard errors clustered by state are presented in parentheses. The number of observations in each subsample is presented in square brackets. * (**) [***] denotes significant at the 0.10 (0.05) [0.01] level in a two-tailed test. Variables are defined in the Appendix.